Can You Deliver AI-Generated Work to a Client? The Difference Between ‘Permitted’ and ‘Safe to Hand Over’

August 5, 2026Updated: August 5, 2026Articles

I recently came across an X post about the rights and responsibilities involved when designers deliver AI-generated work to clients. It raises an important question that extends beyond model training and copyright ownership: what happens if a problem appears after the finished work has already been delivered?

The most useful distinction is this: permission to use works for AI training, permission under an AI service’s terms to use an output commercially, and the ability to hand that output to a client with confidence are three different issues.

At the same time, rights questions depend heavily on the facts. It would be equally inaccurate to say that AI-generated work can never be delivered, or that the creator using AI automatically bears every possible liability. This article does not attempt to reach a legal conclusion for a specific dispute. Instead, it sets out AI Jiten’s practical framework for drawing a sensible line in ordinary production work.

This article provides general information and a practical risk-management perspective. It is not legal advice for any individual project. For major advertising campaigns, long-term brand assets, contractual warranties, or other high-stakes work, consult a qualified lawyer, patent attorney, or trademark professional in the relevant jurisdiction.

Bottom line: focus on what you are promising the client

AI Jiten does not believe creators need to avoid generative AI altogether. It can be extremely useful for planning, rough concepts, composition studies, and organizing ideas or text.

Once a deliverable is going to a client, however, “commercial use is allowed” is not enough. At least four separate layers need to be checked.

Layer to check	Questions to ask
The AI service’s terms	Does your plan allow commercial use? How are inputs, outputs, and training use handled?
Third-party rights	Does the output come too close to an existing work, real person, logo, or character?
Rights you can pass to the client	Can any copyright actually be assigned? Is the client expecting exclusivity or trademark registration?
Your agreement with the client	What does the contract say about AI disclosure, rights warranties, and what happens if a problem is discovered?

The most important practical questions are how long the work will be used, whether exclusivity is part of its value, and how difficult it would be to replace. An internal concept sketch and a logo intended to represent a company for ten years should not go through the same review process.

Permission to train is not permission to use every output

Under Japanese copyright law, Article 30-4 permits certain uses of copyrighted works for information analysis and other purposes that are not aimed at enjoying the thoughts or emotions expressed in the work, subject to statutory limits. This is the provision most often mentioned in Japanese discussions of AI training.

It does not follow that every resulting output is automatically safe to publish, sell, or deliver. Japan’s Agency for Cultural Affairs also separates the development and training stage from the generation and use stage in its materials on AI and copyright.

At the output stage, questions such as similarity and reliance on an existing work are assessed much as they are for human-made work. A broad style, mood, or idea is not automatically protected in the same way as a specific expression. But a close combination of composition, shapes, arrangement, and other distinctive expressive choices may require much more careful review.

Uploading a specific work for the purpose of generating something similar can also raise a separate issue at the input stage. For a beginner-friendly introduction to the Japanese framework, see AI Jiten’s glossary entry on AI and Copyright.

Liability does not automatically fall entirely on the user

In practice, the studio or independent creator who delivered the work may be the first party a client asks for an explanation. The AI provider will not necessarily step in and handle that client conversation. In that practical sense, the warning in the original X post is worth taking seriously.

Legal responsibility is still not determined by a slogan. It can depend on the service terms, the contract between the parties, what was entered into the AI, the checks performed on the output, and how any loss is connected to those actions. In April 2026, Japan’s Ministry of Economy, Trade and Industry published its Guidance on the Interpretation and Application of Civil Liability in the Utilization and Application of AI. The guidance starts from the fact that case law remains limited and examines responsibility according to how AI is used and what role each party plays.

AI Jiten therefore avoids the blanket claim that the user always bears everything. A more useful goal is to build a production process that lets each party explain its role, then confirm that division of responsibility in the contract.

Assess delivery risk by how difficult the work is to replace

Before learning every detail of copyright doctrine, consider what would happen if the output had to be withdrawn or replaced. This provides a practical way to scale the review process.

Use	Examples	AI Jiten’s approach
Internal supporting material	Concept sketches, composition studies, color studies, text organization	AI is often useful here. Still check confidentiality and the rights in any uploaded material
Public work that is easy to replace	Social posts, blog illustrations, short-term campaign assets	Check the terms and similarity, then have a person review and finish the work
Major work intended for long-term use	Advertising key visuals, packaging, book covers	Keep stronger production records and conduct a deeper rights review; seek specialist advice when appropriate
Brand assets whose value depends on exclusivity	Logos, characters, signature brand graphics	Consider using AI only for ideation rather than treating its output as the finished asset

Outputs involving a real person’s face or voice, another company’s logo or product, or a well-known character require extra caution regardless of how replaceable they are. Copyright may be only one issue; privacy, publicity, trademark, and other rights may also be relevant. The applicable rules differ by country.

Three things to establish before production starts

1. Define the deliverable’s use and lifespan

Confirm where the work will appear, how long it will remain in use, and how prominently it will be displayed. The number of media channels, advertising budget, print volume, and countries of use can all affect the consequences if a problem is discovered later.

2. Check the rights and confidentiality of every input

Unreleased client images and personal or confidential information should not automatically be uploaded to a general-purpose AI service. Prefer material you created yourself, material licensed for the intended use, or material whose provenance and permissions can be verified. The client’s policies and your organization’s rules matter as well.

3. Save the applicable service terms

Commercial-use permission is only one part of the terms. Also review how inputs and outputs are handled, whether the provider may use data for training, prohibited uses, infringement-claim procedures, and any indemnity or protection offered under your plan. Terms change, so record the service, plan, and date of review for the project.

Keep a record of what happened during generation

Avoid casually adding specific artist names, titles, or character names to prompts
When using reference images, record their source and license
Save the generation date, service, key instructions, and selected outputs
Document the human decisions used to redesign composition, form, color, and detail
Keep the client’s review and revision history

The purpose is not to prove that you clicked a button many times. The Agency for Cultural Affairs explains that copyrightability of AI-assisted work depends on human creative intent and creative contribution, assessed across the process on a case-by-case basis. Entering one hundred prompts does not automatically create copyright.

Redrawing an AI output by hand does not automatically make it safe either. If protected expression from a source work remains, changing the production technique does not necessarily remove the issue. Human work should be used to redesign the expression for the project, not merely trace the generated result.

A five-point check before delivery

Similarity to existing work: Does the combination of prominent composition, shapes, objects, or colors strongly call a particular work to mind?
People and brands: Does the output unintentionally contain a recognizable person, logo, product, or character?
Service terms: Does your current plan permit this medium, territory, and use?
Rights being promised: Is the client expecting exclusive use, copyright assignment, trademark registration, or another guarantee?
Explanation and records: Can you explain where AI was used, what a person changed, and what checks were completed?

Image search and trademark databases can provide useful leads, but they do not guarantee clearance. For names and logos used in Japan, the Japan Patent Office explains how to conduct a preliminary search through J-PlatPat. Important final decisions should still be reviewed by a qualified trademark professional.

How to explain AI use to a client

Japan does not currently have one simple rule requiring a label for every use of AI. But concealing AI use can become a trust and contract problem when a project restricts AI, or when the deliverable involves warranties, exclusivity, or long-term brand use.

Instead of stopping at “AI was used,” explain the scope:

Which stages used AI, such as ideation, rough concepts, or asset generation
How much of the generated output remains in the final deliverable
What a person reviewed, changed, or rebuilt
What was checked under the service terms and regarding third-party rights
How replacement or reworking would be handled if a problem were discovered

This is not a confession. It is a way to agree on the production process and the scope of the deliverable. Our article on EU AI Act transparency rules and smartphone AI editing also explains why the degree of alteration and the context of use matter when deciding what should be disclosed.

Bring in a specialist for high-risk work

A general checklist should not be the final word for projects such as these:

Logos, characters, or product names intended for registration or exclusive use
Campaigns with substantial media spending or printing costs that would be difficult to reverse
Work involving a real person’s face, voice, or name
Generated content based on a specific existing work
Contracts requiring copyright assignment or a warranty of non-infringement
Work that will be used in several countries

The same separation between training disputes and downstream use appears in music generation. For a concrete example, see our explanation of the German Suno–GEMA ruling and what it means for generated tracks and YouTube background music.

Conclusion: make the process explainable before delivery

There is no one-line answer that makes every AI-generated deliverable legal or illegal, safe or unsafe. The review process needs to scale with the importance of the work.

Separate permission for AI training from permission to use and deliver an output
Do not assume that commercial-use terms guarantee freedom from third-party claims
Increase human redesign and specialist review as work becomes harder to replace or more dependent on exclusivity
Record inputs, generation steps, human revisions, and review decisions
Agree with the client not only that AI was used, but what was checked and what is being promised

AI Jiten’s conclusion is not that creators should stop using AI. Use AI to create faster, then spend some of the saved time checking whether the client can continue using the work with confidence. That is a more realistic boundary for professional generative-AI use.

Sources checked on August 5, 2026

Laws, government guidance, and AI-service terms can change. Check the latest information and seek professional advice when a specific deliverable requires a legal or contractual determination.